THE BRAINTREE APPROACH
Discover, Evaluate, Deploy, Build.
A chatbot here and a point tool there can leave the business carrying costs without a clear owner. Braintree’s AI Value Framework tests each proposed use case before deployment. The decision includes the data, the controls and the work after launch.
01
Discover
Map the operations and workflows where an assistant or agent could make a meaningful difference. Start with the task, its users and the outcome the business needs.
02
Evaluate
Model total cost of ownership and expected value. Present the case to management, including data readiness, governance and the people who will own the result before deployment is approved.
03
Deploy
Configure a supported Microsoft or Azure capability where it fits the job. Ground it in the permitted information, test its behaviour and establish the review and operating responsibilities.
04
Build
Assess a custom agent when the requirement is specific to your business. Apply content safety, evaluations and a logged audit trail, with the build and handover scoped to that use case.
SERVICE SCOPE
What’s in scope.
01
Use-case discovery
Map repetitive work across operations, documents, customer service, finance and internal knowledge. Identify the task owner, current effort, source systems and points where human judgement is needed. Prioritise a use case that can be evaluated against a clear baseline.
02
Copilot Studio agents
Evaluate low-code agents that use approved organisational knowledge and connect to defined business processes. Establish what an agent can read, what it can change and when a person must review an answer or action.
03
Azure AI and Microsoft Foundry
Use Azure AI services and Microsoft Foundry where the use case needs a custom application or agent. The technical scope can include model selection, knowledge grounding, evaluation sets, content safety and logged activity.
04
Knowledge and document workflows
Explore assistants that find policy information, draft service responses, extract document content or prepare reporting commentary. Tie each output to the relevant source and review process so users can assess the answer in context.
Make the decision before the deployment.
Braintree’s published AI Value Framework moves through discovery, evaluation, deployment and custom build where required. The useful output is a management decision about one defined task, with the costs, risks and responsibilities visible.
✓
A prioritised use-case register with a named business owner.
✓
A data and access map, including sources the agent may use.
✓
An evaluation plan, success measures and human approval points.
✓
A TCO model that includes licensing, consumption, integration and ongoing operation.
Choose the right implementation route.
An existing Microsoft capability may meet the requirement; a custom agent is appropriate when the task, knowledge or process is specific to the business. Microsoft 365 Copilot adoption belongs with the workplace environment. An agent acting on customer or finance records also needs the relevant Dynamics process and permissions to be ready.
01
Choose work that can be evaluated
Repeatedly searching policy documents, extracting information from incoming files or preparing a first draft can provide a bounded starting point. Begin with the task, its users and the decision that follows the output. A general request to “add AI” does not reveal which information is authoritative or what a useful result looks like.
Azure AI services and Copilot Studio support different application and agent patterns. The choice depends on the knowledge source, required actions, integration and operating constraints. An assistant that retrieves information has a different risk and permission profile from an agent that updates an order. The initial scope should identify that distinction before selecting a model or connector.
02
Prepare an evaluation set before a production rollout
Bring representative inputs, expected answers or acceptance criteria, known failure examples and the people qualified to judge the result. Include ambiguous, incomplete and out-of-scope requests. Microsoft’s evaluation guidance distinguishes output quality from the behaviour of an agent’s workflow; both matter when tools or external systems are involved.
Agree which data may be indexed, who can retrieve it, how source material is refreshed and when a human must approve an action. The customer supplies subject expertise and accepts business suitability. The delivery scope should identify integration, evaluation, deployment and operational handover work separately. A useful handover records the prompts or configuration, test results, known limits, monitoring needs and the owner responsible for revisiting them when the process changes.
03
Decisions that change the AI design
Decision
Evidence to examine
What to agree
Knowledge assistant or process agent
Whether the user needs an answer, a draft or a change to a business record.
Permitted tools and actions, approval points and audit requirements.
Existing capability or custom build
Fit with supported Microsoft functions and required business context.
The integration and maintenance burden justified by the use case.
Pilot acceptance
Representative tasks, failure costs and measured operating consumption.
The evidence required to expand, revise or stop the experiment.
BEFORE YOU DECIDE
Practical questions.
▸
Is a chatbot the default deliverable?
▸
Can an agent act without review?
▸
What determines data location?
▸
How do we assess answers that sound convincing but are wrong?
▸
Can an agent use SharePoint knowledge?
▸
What contributes to running cost?
▸
Who owns the assistant after launch?
BUILD YOUR BUSINESS CASE
Propose an AI use case with a measurable job to do.
The problem to solve.
An impressive demonstration does not establish whether a model is accurate enough, whether the right data is available or whether people can rely on the output.
What success should look like.
Define one bounded use case, a human review path and measurable acceptance criteria before committing to wider deployment.
Bring these into the discussion.
A useful starting point for your team and the people approving the investment.
Explore the connected technology.
Useful reading for your proposal.
Evidence, examples and questions to take into the decision.
Customer story
27 November 2025
Customer story
Bioforum: AI with validation and human oversight
Follow a clinical-research use case where trustworthy outputs were central to the design.
Read the customer story
Insight
13 April 2026
AI planning
Start with an AI strategy, not a shopping list
Frame the business problem, governance and adoption before selecting tools.
Read the insight
Insight
20 July 2026
Data governance
Govern AI access to operational data
Use the SQL example to sharpen questions about what the assistant may see and do.
Read the insight
LET’S GROW.
Work through the next step with Braintree.
Bring a useful task and representative examples. We can help define a pilot that answers a real investment question.
